Where should external non-PyPI dependencies be located?

[matthewfeickert]

In the installation docs it is noted that the cross section computers require additional external dependencies.

smodels/docs/manual/source/Installation.rst

Lines 25 to 32 in 022a2a3

	In addition, the :ref:`cross section computers <xsecCalc>` provided by :ref:`smodelsTools.py <smodelsTools>`
	require:
	* `Pythia 8.3 <https://arxiv.org/abs/1410.3012>`_ (needs a C++ compiler) or `Pythia 6.4.27 <http://arxiv.org/abs/hep-ph/0603175>`_ (needs gfortran)
	* `NLL-fast <http://pauli.uni-muenster.de/~akule_01/nllwiki/index.php/NLL-fast>`_ 1.2 (7 TeV), 2.1 (8 TeV), and 3.1 (13 TeV) (needs a Fortran compiler)
	* `Resummino <https://resummino.hepforge.org>`_ (needs a C++ compiler and gfortran). Moreover, in rpm-based Linux distributions, this tool needs boost, boost-devel, gsl and gsl-devel. For deb-based distributions, libboost-dev and libgsl-dev are required.
	The tools themselves, i.e. Pythia6|8, NLL-fast, and/or Resummino need not be installed separately, as the SModelS build system takes care of that (see below). SModelS however expects the tools' dependencies (boost, gsl for Resummino, as well as the compilers) to be installed by the user.

If a user wanted to install these additional dependencies themselves (e.g. getting Pythia8 from conda-forge) where does smodels look for them?

e.g. In the example of getting Pythia8 from conda-forge, libpythia8.so/libpythia8.dylib would be installed under $CONDA_PREFIX/lib/. Does smodels need this to be symlinked then under some specific location? Or is there a way to tell smodels where to look?

cc @WolfgangWaltenberger as he wrote most of the installation docs on this section in a55ccb5.

[WolfgangWaltenberger]

Right, so far we had not really foreseen such usages. Currently one would need to install from source, then e.g. in the top level directory execute: "make pythia8 PYTHIA8HOME=DIR PYTHIAVER=VER" -- not very elegant, I know, see https://github.com/SModelS/smodels/blob/main/smodels/lib/pythia8/Makefile
I guess we could check if CONDA_PREFIX is defined, and if it is, see if we can use it? (Doesnt sound completely safe though). Or asked differently, what would be the conda-correct way of doing this? (Should be noted though that we consider our cross section computers to be optional, not "core" SModelS)

Wolfgang

[matthewfeickert]

Right, so far we had not really foreseen such usages. Currently one would need to install from source, then e.g. in the top level directory execute: "make pythia8 PYTHIA8HOME=DIR PYTHIAVER=VER" -- not very elegant, I know, see https://github.com/SModelS/smodels/blob/main/smodels/lib/pythia8/Makefile

@WolfgangWaltenberger Overriding the make arguments in

smodels/smodels/lib/pythia8/Makefile

Lines 7 to 12 in 022a2a3

	PYTHIAVER := $(shell ./installer.py -v)
	PYTHIA8HOME := pythia$(PYTHIAVER)/
	CXXFLAGS := -O3 -I$(PYTHIA8HOME)/include -I$(PYTHIA8HOME)/include/Pythia8/ -std=c++14
	LDFLAGS := -L$(PYTHIA8HOME)/lib/ -L$(PYTHIA8HOME)/lib -Wl,-rpath,$(PWD)/$(PYTHIA8HOME)/lib
	XMLDOC := $(PYTHIA8HOME)/share/Pythia8/xmldoc

is fine.

I guess we could check if CONDA_PREFIX is defined, and if it is, see if we can use it? (Doesnt sound completely safe though).

I think to support external installs more generally one would allow the user to pass the root of a directory tree and then before attempting to do anything you'd check to see if all necessary directories and files are present.

Or asked differently, what would be the conda-correct way of doing this?

The conda-forge way of doing things would to just be relying on other conda-forge packages to exist, but those again are just going to create a standard bin, lib, include, share, man, etc directory tree in $CONDA_PREFIX and then install things in there as if it was a typical filesystem.

(Should be noted though that we consider our cross section computers to be optional, not "core" SModelS)

Ah okay that's very good to know. So if someone just installed the smodels Python package from PyPI and stopped that wouldn't be viewed in most cases as an incomplete smodels install?

[WolfgangWaltenberger]

Yes that is correct. "pip install smodels" and just not compiling any interfaces for the cross section computers would not be considered an incomplete install by SModelS. The idea is that SModelS reads in the cross sections from eg the slha files, and we do not care where they come from. Its just for our own use cases and for the most prominent BSM models we wrapped the most obvious tools (pythia, nllfast, resummino), and shipped this with SModelS. In principle we could even turn this into its own independent package/tool (except our human ressources are limited and I would hate the idea of having to maintain yet another separate codebase).

[matthewfeickert]

Yes that is correct. "pip install smodels" and just not compiling any interfaces for the cross section computers would not be considered an incomplete install by SModelS. The idea is that SModelS reads in the cross sections from eg the slha files, and we do not care where they come from.

Great. Thanks for this info, @WolfgangWaltenberger!

Its just for our own use cases and for the most prominent BSM models we wrapped the most obvious tools (pythia, nllfast, resummino), and shipped this with SModelS. In principle we could even turn this into its own independent package/tool (except our human resources are limited and I would hate the idea of having to maintain yet another separate codebase).

Sounds good. In coming weeks I'll give this a look as to how to get externally installed tools to work here. Though, I am also going to be time limited, so it might take some time. :)